pdf_parser.py

"""
pdf_parser.py
─────────────
Utility called by fastapi_main.py before invoking the LangGraph pipeline.

Accepts a list of (filename, raw_bytes) tuples, parses each PDF with LiteParse,
and returns a list of {"filename": str, "content": str} dicts ready to be stored
in the LangGraph state as `pdf_context`.

Each file's extracted text is truncated to MAX_SOURCE_CHARS (same cap used for
web-scraped content) so the synthesis token budget is not blown.
"""

import tempfile
import os
from liteparse import LiteParse
from config import MAX_SOURCE_CHARS


def parse_pdfs(file_bytes_list: list[tuple[str, bytes]]) -> list[dict] | None:
    """
    Parse a list of PDF files and return their text content.

    Args:
        file_bytes_list: List of (filename, raw_bytes) tuples.

    Returns:
        List of {"filename": str, "content": str} dicts, or None if input is empty.
        Files that fail to parse are skipped with a warning printed to stdout.
    """
    if not file_bytes_list:
        return None

    parser = LiteParse()
    results = []

    for filename, raw_bytes in file_bytes_list:
        # Write to a temp file because LiteParse expects a file path
        suffix = os.path.splitext(filename)[-1] or ".pdf"
        tmp = None
        try:
            with tempfile.NamedTemporaryFile(suffix=suffix, delete=False) as tmp:
                tmp.write(raw_bytes)
                tmp_path = tmp.name

            print(f"   [PDF] Parsing: {filename} ({len(raw_bytes) / 1024:.1f} KB)")
            parsed = parser.parse(tmp_path)

            # Concatenate all page texts
            full_text = "\n\n".join(
                page.text for page in parsed.pages if page.text and page.text.strip()
            )

            if not full_text.strip():
                print(f"   ⚠️ [PDF] No text extracted from: {filename} (may be image-only)")
                continue

            # Truncate to the same per-source cap used for web content
            if len(full_text) > MAX_SOURCE_CHARS:
                marker = "\n\n[PDF content truncated to fit the source budget.]"
                full_text = full_text[: MAX_SOURCE_CHARS - len(marker)].rsplit(" ", 1)[0] + marker

            results.append({"filename": filename, "content": full_text})
            print(f"   ✅ [PDF] Parsed: {filename} ({len(full_text):,} chars extracted)")

        except Exception as exc:
            print(f"   ⚠️ [PDF] Failed to parse '{filename}': {type(exc).__name__}: {exc}")

        finally:
            # Always clean up the temp file
            if tmp is not None and os.path.exists(tmp.name):
                try:
                    os.unlink(tmp.name)
                except OSError:
                    pass

    return results if results else None